Out-of-Distribution Generalization on GPQA Diamond OOD
71.5Sav.TTT no-QK
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TTT no-QKLabeling protocol=Supervised labels, delta=0.1, zero-shot=true2026.04 | 71.5 | 30 | |
| TTT QK (dh=128)Labeling protocol=Supervised labels, delta=0.1, zero-shot=true2026.04 | 66.5 | 21 | |
| TTT QK (dh=128)Labeling protocol=Consistent labels (no ground truth), delta=0.1, zero-shot=true2026.04 | 65.3 | 32.8 | |
| Static ProbeLabeling protocol=Supervised labels, delta=0.1, zero-shot=true2026.04 | 64.3 | 27 | |
| Static ProbeLabeling protocol=Consistent labels (no ground truth), delta=0.1, zero-shot=true2026.04 | 60.2 | 32.8 | |
| TTT no-QKLabeling protocol=Consistent labels (no ground truth), delta=0.1, zero-shot=true2026.04 | 59.8 | 31.8 |